FastConcat3.diff

   1 From: Tony Balinskt <ajbj@free.fr>
   2 Subject: Perform faster macro concatenation
   3
   4 Available as a patch:
   5
   6         http://sourceforge.net/tracker/index.php?func=detail&aid=971477&group_id=11005&atid=311005
   7         [ 971477 ] Perform faster macro concatenation
   8         FastConcat.diff 2004-06-11 15:55
   9
  10 Previously, concatenation was treated as a binary operator, taking two values
  11 from the stack and concatenating them. Where an expression involves more than
  12 one concatenation, this involved creating temporary strings, requiring
  13 separate allocation, which would only be discarded later, for each application
  14 of the binary concatenation. To make matters worse, concatenation would rescan
  15 each string to determine its length using strlen().
  16
  17 This modification changes the way the concatenation operator is handled.
  18 Instead of treating it as a binary operator, it sees it as a list operator,
  19 rather like the comma in an argument list. Thus a concatenation expression
  20 is treated as a counted sequence of subexpressions. The count is coded into
  21 the "bytecode" to tell the concat() function (in interpret.c) how many
  22 expression values to pull off the stack. The function then exploits the
  23 fact that string DataValue structures now include the strings' lengths to
  24 work out the total allocation length required for the whole result. (This
  25 is based on the previous implementation of makeArrayKeyFromArgs().)
  26
  27 I have also added static functions longAsStr() and lenLongAsStr() to generate
  28 a string representation of a number, or just to work out its length. This
  29 allows us to avoid calling sprintf() with "%d" all the time. The new
  30 AllocStringOfNumber() exploits this, and is used in various places in
  31 interpret.c. Note that longAsStr() returns the string value in a static buffer
  32 which must be used/copied immediately to avoid a second call overwriting
  33 its content.
  34
  35 ---
  36
  37  source/interpret.c |  228 ++++++++++++++++++++++++++++++++++++++---------------
  38  source/interpret.h |    1
  39  source/parse.y     |   15 ++-
  40  3 files changed, 178 insertions(+), 66 deletions(-)
  41
  42 diff --quilt old/source/interpret.c new/source/interpret.c
  43 --- old/source/interpret.c
  44 +++ new/source/interpret.c
  45 @@ -83,10 +83,12 @@ static void restoreContext(RestartData *
  46  #define OP(name, fn) static int fn(void);
  47  #include "ops.h"
  48  #undef OP
  49  static int returnValOrNone(int valOnStack);
  50
  51 +static int concatenateNwithSep(int nVals, const char *sep, char **result,
  52 +        int leaveParams);
  53  static void freeSymbolTable(Symbol *symTab);
  54  static int errCheck(const char *s);
  55  static int execError(const char *s1, const char *s2);
  56  static rbTreeNode *arrayEmptyAllocator(void);
  57  static rbTreeNode *arrayAllocateNode(rbTreeNode *src);
  58 @@ -791,10 +793,64 @@ Symbol *PromoteToGlobal(Symbol *sym)
  59
  60      return sym;
  61  }
  62
  63  /*
  64 +** Convert a long value to its decimal string representation, returned in a
  65 +** static string.
  66 +*/
  67 +const char *longAsStr(long val)
  68 +{
  69 +    static const char digits[] = "0123456789";
  70 +    static char res[TYPE_INT_STR_SIZE(val) + 1];
  71 +    char *pos = &res[TYPE_INT_STR_SIZE(val)];
  72 +
  73 +    /* the string is built backwards, so start by null terminating it */
  74 +    *pos = 0;
  75 +
  76 +    if (val >= 0) {
  77 +        /* do-while loop will deal with the val == 0 case */
  78 +        do {
  79 +            /* we can use the modulo (%) operator here */
  80 +            *--pos = digits[val % 10];
  81 +            val /= 10;
  82 +        } while (val != 0);
  83 +    }
  84 +    else {
  85 +        /* we don't use the modulo (%) operator since its behaviour with
  86 +           negative numbers is undefined by the C standards */
  87 +        do {
  88 +            long val10 = val / 10;
  89 +            *--pos = digits[(10 * val10) - val];
  90 +            val = val10;
  91 +        } while (val != 0);
  92 +        *--pos = '-';
  93 +    }
  94 +    return pos;
  95 +}
  96 +
  97 +/*
  98 +** Calculate the length of the decimal string representation of a long value.
  99 +*/
 100 +int lenLongAsStr(long val)
 101 +{
 102 +    int len = 0;
 103 +
 104 +    if (val < 0) {
 105 +        len++;  /* for leading '-' */
 106 +    }
 107 +
 108 +    /* do-while loop will deal with the val == 0 case */
 109 +    do {
 110 +        len++;
 111 +        val /= 10;
 112 +    } while (val != 0);
 113 +
 114 +    return len;
 115 +}
 116 +
 117 +/*
 118  ** Allocate memory for a string, and keep track of it, such that it
 119  ** can be recovered later using GarbageCollectStrings.  (A linked list
 120  ** of pointers is maintained by threading through the memory behind
 121  ** the returned pointers).  Length does not include the terminating null
 122  ** character, so to allocate space for a string of strlen == n, you must
 123 @@ -881,10 +937,16 @@ int AllocNStringNCpy(NString *string, co
 124  char *AllocStringCpy(const char *s)
 125  {
 126      return AllocStringNCpy(s, s ? strlen(s) : 0);
 127  }
 128
 129 +/* Allocate a string holding the decimal value of the number */
 130 +char *AllocStringOfNumber(long val)
 131 +{
 132 +    return AllocStringCpy(longAsStr(val));
 133 +}
 134 +
 135  /*
 136   * Allocate a new NString buffer, containing a copy of the given string.
 137   * The length is set to the length of the string and resulting string is
 138   * guaranteed to be \0-terminated.
 139   */
 140 @@ -1120,12 +1182,11 @@ static void freeSymbolTable(Symbol *symT
 141      do { \
 142          char *__str; \
 143          if (dataVal.tag == STRING_TAG) { \
 144              __str = dataVal.val.str.rep; \
 145          } else if (dataVal.tag == INT_TAG) { \
 146 -            __str = AllocString(TYPE_INT_STR_SIZE(int)); \
 147 -            sprintf(__str, "%d", dataVal.val.n); \
 148 +            __str = AllocStringOfNumber(dataVal.val.n); \
 149          } else { \
 150              return(execError("incompatible type in string context: %s", \
 151                      tagToStr(dataVal.tag))); \
 152          } \
 153          string = __str; \
 154 @@ -1281,13 +1342,12 @@ static int pushArgVal(void)
 155
 156      POP_INT(argNum);
 157      --argNum;
 158      nArgs = FP_GET_ARG_COUNT(FrameP);
 159      if (argNum >= nArgs || argNum < 0) {
 160 -        char argStr[TYPE_INT_STR_SIZE(argNum)];
 161 -        sprintf(argStr, "%d", argNum + 1);
 162 -       return execError("referenced undefined argument: $args[%s]", argStr);
 163 +        return execError("referenced undefined argument: $args[%s]",
 164 +                         longAsStr(argNum + 1));
 165      }
 166      PUSH(FP_GET_ARG_N(FrameP, argNum));
 167      return STAT_OK;
 168  }
 169
 170 @@ -1313,15 +1373,13 @@ static int pushArgArray(void)
 171      if (resultArray->tag != ARRAY_TAG) {
 172          resultArray->tag = ARRAY_TAG;
 173          resultArray->val.arrayPtr = ArrayNew();
 174
 175          for (argNum = 0; argNum < nArgs; ++argNum) {
 176 -            char intStr[TYPE_INT_STR_SIZE(argNum)];
 177 -
 178 -            sprintf(intStr, "%d", argNum + 1);
 179              argVal = FP_GET_ARG_N(FrameP, argNum);
 180 -            if (!ArrayInsert(resultArray, AllocStringCpy(intStr), &argVal)) {
 181 +            if (!ArrayInsert(resultArray, AllocStringOfNumber(argNum + 1),
 182 +                             &argVal)) {
 183                  return(execError("array insertion failure", NULL));
 184              }
 185          }
 186      }
 187      PUSH(*resultArray);
 188 @@ -1907,30 +1965,106 @@ static int power(void)
 189      PUSH_INT(n3);
 190      return errCheck("exponentiation");
 191  }
 192
 193  /*
 194 -** concatenate two top items on the stack
 195 -** Before: TheStack-> str2, str1, next, ...
 196 -** After:  TheStack-> result, next, ...
 197 +** A helper routine used in concat(), and makeArrayKeyFromArgs().
 198 +** Concatenate a number of values from the stack and return the result as a
 199 +** character pointer in *result and its length as the return value, or less
 200 +** than zero on failure. If a divider is specified, add it between each of the
 201 +** stack elements. The stack elements are popped from the stack if leaveParams
 202 +** is false.
 203 +*/
 204 +static int concatenateNwithSep(int nVals, const char *sep, char **result,
 205 +        int leaveParams)
 206 +{
 207 +    DataValue value;
 208 +    char *res = NULL;
 209 +    char *pos;
 210 +    int len;
 211 +    int sepLen;
 212 +    int i;
 213 +
 214 +    *result = NULL;
 215 +
 216 +    if (sep == NULL) {
 217 +        sep = "";
 218 +    }
 219 +    sepLen = strlen(sep);
 220 +
 221 +    /* evaluate total length (upper limit) */
 222 +    len = sepLen * (nVals - 1);
 223 +    for (i = nVals - 1; i >= 0; --i) {
 224 +        PEEK(value, i);
 225 +        if (value.tag == INT_TAG) {
 226 +            len += lenLongAsStr(value.val.n);
 227 +        }
 228 +        else if (value.tag == STRING_TAG) {
 229 +            len += value.val.str.len;
 230 +        }
 231 +        else {
 232 +            return -1;  /* invalid type */
 233 +        }
 234 +    }
 235 +
 236 +    /* allocate the string */
 237 +    res = AllocString(len + 1);
 238 +    pos = res;
 239 +    /* write everything into the result */
 240 +    for (i = nVals - 1; i >= 0; --i) {
 241 +        PEEK(value, i);
 242 +        if (value.tag == INT_TAG) {
 243 +            pos += strlen(strcpy(pos, longAsStr(value.val.n)));
 244 +        }
 245 +        else { /* value.tag == STRING_TAG */
 246 +            strcpy(pos, value.val.str.rep);
 247 +            pos += value.val.str.len;
 248 +        }
 249 +        if (i && sepLen) {
 250 +            strcpy(pos, sep);
 251 +            pos += sepLen;
 252 +        }
 253 +    }
 254 +
 255 +    /* remove the source expression values */
 256 +    if (!leaveParams) {
 257 +        while (nVals--) {
 258 +            POP(value);
 259 +        }
 260 +    }
 261 +
 262 +    /* now return the results */
 263 +    *result = res;
 264 +    return pos - res;
 265 +}
 266 +
 267 +/*
 268 +** concatenate a number of strings and push the result onto the stack
 269 +**
 270 +** Before: Prog->  [nExpr], next, ...
 271 +**         TheStack-> exprValN, ... exprVal1, next, ...
 272 +** After:  Prog->  nExpr, [next], ...
 273 +**         TheStack-> concatResult, next, ...
 274  */
 275  static int concat(void)
 276  {
 277 -    char *s1, *s2, *out;
 278 -    int len1, len2;
 279 +    char *out;
 280 +    int len;
 281
 282 -    DISASM_RT(PC-1, 1);
 283 -    STACKDUMP(2, 3);
 284 +    int nExpr;
 285 +
 286 +    nExpr = PC->value;
 287 +    PC++;
 288 +
 289 +    DISASM_RT(PC-2, 2);
 290 +    STACKDUMP(nExpr, 3);
 291
 292 -    POP_STRING(s2);
 293 -    POP_STRING(s1);
 294 -    len1 = strlen(s1);
 295 -    len2 = strlen(s2);
 296 -    out = AllocString(len1 + len2 + 1);
 297 -    strncpy(out, s1, len1);
 298 -    strcpy(&out[len1], s2);
 299 -    PUSH_STRING(out, len1 + len2);
 300 +    len = concatenateNwithSep(nExpr, "", &out, False);
 301 +    if (len < 0) {
 302 +        return(execError("can only concatenate with string or integer", NULL));
 303 +    }
 304 +    PUSH_STRING(out, len);
 305      return STAT_OK;
 306  }
 307
 308  /*
 309  ** Call a subroutine or function (user defined or built-in).  Args are the
 310 @@ -2258,49 +2392,15 @@ int ArrayCopy(DataValue *dstArray, DataV
 311  ** I really need to optimize the size approximation rather than assuming
 312  ** a worst case size for every integer argument
 313  */
 314  static int makeArrayKeyFromArgs(int nArgs, char **keyString, int leaveParams)
 315  {
 316 -    DataValue tmpVal;
 317 -    int sepLen = strlen(ARRAY_DIM_SEP);
 318 -    int keyLength = 0;
 319 -    int i;
 320 +    int len;
 321
 322 -    keyLength = sepLen * (nArgs - 1);
 323 -    for (i = nArgs - 1; i >= 0; --i) {
 324 -        PEEK(tmpVal, i);
 325 -        if (tmpVal.tag == INT_TAG) {
 326 -            keyLength += TYPE_INT_STR_SIZE(tmpVal.val.n);
 327 -        }
 328 -        else if (tmpVal.tag == STRING_TAG) {
 329 -            keyLength += tmpVal.val.str.len;
 330 -        }
 331 -        else {
 332 -            return(execError("can only index array with string or int.", NULL));
 333 -        }
 334 -    }
 335 -    *keyString = AllocString(keyLength + 1);
 336 -    (*keyString)[0] = 0;
 337 -    for (i = nArgs - 1; i >= 0; --i) {
 338 -        if (i != nArgs - 1) {
 339 -            strcat(*keyString, ARRAY_DIM_SEP);
 340 -        }
 341 -        PEEK(tmpVal, i);
 342 -        if (tmpVal.tag == INT_TAG) {
 343 -            sprintf(&((*keyString)[strlen(*keyString)]), "%d", tmpVal.val.n);
 344 -        }
 345 -        else if (tmpVal.tag == STRING_TAG) {
 346 -            strcat(*keyString, tmpVal.val.str.rep);
 347 -        }
 348 -        else {
 349 -            return(execError("can only index array with string or int.", NULL));
 350 -        }
 351 -    }
 352 -    if (!leaveParams) {
 353 -        for (i = nArgs - 1; i >= 0; --i) {
 354 -            POP(tmpVal);
 355 -        }
 356 +    len = concatenateNwithSep(nArgs, ARRAY_DIM_SEP, keyString, leaveParams);
 357 +    if (len < 0) {
 358 +      return(execError("can only index array with string or int.", NULL));
 359      }
 360      return(STAT_OK);
 361  }
 362
 363  /*
 364 @@ -3007,10 +3107,14 @@ static void disasm(Inst *inst, int nInst
 365                          j == OP_BRANCH_NEVER || j == OP_BRANCH_TRUE) {
 366                      printf("to=(%d) %p", inst[i+1].value,
 367                              &inst[i+1] + inst[i+1].value);
 368                      ++i;
 369                  }
 370 +                else if (j == OP_CONCAT) {
 371 +                    printd("nExpr=%d", inst[i+1].value);
 372 +                    ++i;
 373 +                }
 374                  else if (j == OP_SUBR_CALL) {
 375                      printf("%s (%d arg)", inst[i+1].sym->name, inst[i+2].value);
 376                      i += 2;
 377                  }
 378                  else if (j == OP_BEGIN_ARRAY_ITER) {
 379 diff --quilt old/source/interpret.h new/source/interpret.h
 380 --- old/source/interpret.h
 381 +++ new/source/interpret.h
 382 @@ -170,10 +170,11 @@ int ContinueMacro(RestartData *continuat
 383  void RunMacroAsSubrCall(Program *prog);
 384  void PreemptMacro(void);
 385  char *AllocString(int length);
 386  char *AllocStringNCpy(const char *s, int length);
 387  char *AllocStringCpy(const char *s);
 388 +char *AllocStringOfNumber(long val);
 389  int AllocNString(NString *string, int length);
 390  int AllocNStringNCpy(NString *string, const char *s, int length);
 391  int AllocNStringCpy(NString *string, const char *s);
 392  void GarbageCollectStrings(void);
 393  void FreeRestartData(RestartData *context);
 394 diff --quilt old/source/parse.y new/source/parse.y
 395 --- old/source/parse.y
 396 +++ new/source/parse.y
 397 @@ -88,11 +88,11 @@ static int AllowDefine;
 398  %token <sym> STRING SYMBOL
 399  %token <num> NUMBER
 400  %token DELETE ARG_LOOKUP
 401  %token IF WHILE ELSE FOR BREAK CONTINUE RETURN
 402  %token <acc> DEFINE
 403 -%type <num> arglist
 404 +%type <num> arglist catlist
 405  %type <inst> cond comastmts for while else and or arrayexpr
 406  %type <sym> evalsym
 407
 408  %nonassoc IF_NO_ELSE
 409  %nonassoc ELSE
 410 @@ -346,13 +346,20 @@ arglist:    /* nothing */ {
 411              }
 412              | arglist ',' expr {
 413                  $$ = $1 + 1;
 414              }
 415              ;
 416 -expr:       numexpr %prec CONCAT
 417 -            | expr numexpr %prec CONCAT {
 418 -                ADD_OP(OP_CONCAT);
 419 +catlist:    numexpr %prec CONCAT {
 420 +                $$ = 1;
 421 +            }
 422 +            | catlist numexpr %prec CONCAT {
 423 +                $$ = $1 + 1;
 424 +            }
 425 +expr:       catlist {
 426 +                if ($1 > 1) {
 427 +                    ADD_OP(OP_CONCAT); ADD_IMMED($1);
 428 +                }
 429              }
 430              ;
 431  initarraylv:    SYMBOL {
 432                      ADD_OP(OP_PUSH_ARRAY_SYM); ADD_SYM($1); ADD_IMMED(1);
 433                  }